High-concordance detection of EGFR mutations across diverse biospecimens in a large-scale NSCLC cohort using droplet digital PCR and next-generation sequencing

Background

Non-small cell lung cancer (NSCLC) is the most prevalent subtype of lung cancer worldwide, accounting for approximately 85% of all lung cancer cases and representing a leading cause of cancer-related mortality globally. Somatic mutations in the epidermal growth factor receptor (EGFR) gene—particularly exon 19 deletions (Ex19del), the L858R point mutation, and the T790M resistance mutation—are critical drivers in NSCLC and key determinants of response to EGFR tyrosine kinase inhibitors (TKIs). Accurate detection and quantification of these mutations are essential for molecular stratification and individualized treatment selection.

While next-generation sequencing (NGS) enables comprehensive genomic profiling, its analytical sensitivity is constrained by sequencing depth, background noise, and error correction strategies, particularly in samples with limited tumor DNA content. Droplet digital PCR (ddPCR) offers a complementary approach by partitioning DNA templates into tens of thousands of nanoliter-scale droplets, enabling absolute quantification at the single-molecule level with reported limits of detection reaching 0.01% VAF under optimized conditions. However, large-scale systematic evaluations of ddPCR performance across diverse clinical specimen types—and its quantitative concordance with NGS—have remained limited.

Study Design and Methods

This retrospective multi-center study analyzed 936 analytical mutation-target records from 747 NSCLC patients with NGS-detected EGFR mutations. The records represented native DNA or corresponding pre-capture NGS-library derivatives from four specimen types: fresh tumor tissue (ttDNA), formalin-fixed paraffin-embedded tissue (ffpeDNA), plasma-derived cell-free DNA (cfDNA), and pleural effusion supernatant DNA (peDNA). A total of 177 patients contributed more than one analytical record, and 37 contributed more than one source biospecimen.

ddPCR reactions were designed to detect EGFR L858R (c.2573T>G), Ex19del, and T790M (c.2369C>T) using the TargetingOne D50 Automatic Digital PCR system, which partitions each sample into approximately 50,000–60,000 nanoliter-scale droplets. ddPCR results were compared with NGS for mutation detection and variant allele frequency (VAF) quantification using correlation analysis, Bland-Altman analysis, and positive percent agreement (PPA) with exact 95% confidence intervals. To account for within-patient and within-biospecimen clustering, cluster bootstrap methods (1,000 resamples, clustered by patient ID) and repeated-measures Bland-Altman approaches were applied.

Key Findings

Overall Concordance: Across all 936 analytical records, ddPCR demonstrated a high overall PPA with NGS of 98.82% (925/936; exact 95% CI, 97.91%–99.41%). Mutation-specific PPAs were 98.93% for L858R (370/374; exact 95% CI, 97.28%–99.71%), 99.48% for Ex19del (385/387; exact 95% CI, 98.15%–99.94%), and 97.14% for T790M (170/175; exact 95% CI, 93.46%–99.07%).

VAF Correlation: VAFs measured by ddPCR and NGS were strongly correlated (Pearson's r = 0.975). Bland-Altman analysis revealed a mean difference of 3.18% (95% CI: 2.84%–3.52%), indicating that ddPCR tended to yield slightly higher VAF estimates than NGS on average, with limits of agreement ranging from –7.29% to 13.65%.

VAF-Stratified Performance: Inter-method differences were minimal in the low-VAF group (<1%; mean difference 0.03%, SD 0.28%), increased moderately in the intermediate-VAF group (1–10%; mean difference 0.75%, SD 1.66%), and were broader in the high-VAF group (≥10%; mean difference 5.19%, SD 6.26%). These findings suggest that quantitative divergence becomes more evident as mutant allele abundance increases, likely reflecting differences in amplification efficiency, probe-based signal detection, and sequencing-based representation rather than reduced sensitivity alone.

Mutation-Type Stratification: L858R mutations showed near-perfect correlation (r = 0.992, MAE = 2.24%), while Ex19del mutations exhibited greater variability (r = 0.972, MAE = 6.40%)—likely due to the heterogeneous variable-length indel nature of exon 19 deletions, which pose challenges for short-read NGS alignment and quantification. T790M mutations also showed high concordance (r = 0.978, MAE = 1.55%).

Sample-Type Stratification: Plasma-derived cfDNA showed the largest mean bias (4.43%) and widest limits of agreement (–10.84% to 19.69%), reflecting the combined effects of low tumor fraction and high background cfDNA. In contrast, pleural effusion supernatant DNA (peDNA) demonstrated the smallest bias (mean 2.19%) and narrowest limits of agreement, consistent with previous findings that pleural effusion can enrich tumor DNA and yield more stable mutation signals than plasma. FFPE and fresh tumor tissue DNA exhibited intermediate variability with overall good concordance.

Pre-Capture Libraries as Surrogate Templates: Paired analyses comparing native DNA with corresponding pre-capture NGS libraries showed strong correlations for cfDNA libraries (47 pairs; r = 0.994; mean difference –0.72%), tumor tissue-derived L858R libraries (46 pairs; r = 0.993), and Ex19del libraries (47 pairs; r = 0.993). These results indicate that pre-capture library DNA does not substantially alter detection outcomes, supporting the feasibility of using archived libraries as alternative templates for ddPCR in low-input scenarios.

Clinical Significance

ddPCR serves as both an effective complement to NGS and a dependable alternative in specific clinical scenarios, particularly for rapid testing, low-frequency mutation detection, and repeated measurements across diverse sample types—providing robust support for clinical diagnosis and precision oncology decision-making.

The validated use of pre-capture NGS libraries as ddPCR templates offers a practical solution when original DNA is insufficient, potentially reducing the need for repeated biopsies and enabling more efficient use of limited clinical samples.

Limitations

The low-VAF reference-standard experiment was exploratory: all 10 replicates at the nominal 0.01% level were detected using 100 ng input, but the exact two-sided 95% CI for 10/10 detection was 69.15%–100%, and no formal limit of blank (LoB), limit of detection (LoD), or limit of quantification (LoQ) validation was performed. Consequently, 0.01% should not be interpreted as a general analytical LoD, especially for low-input native plasma cfDNA. Additionally, because only NGS-confirmed EGFR-positive samples were included, specificity and negative percent agreement could not be assessed. The study should be interpreted as an evaluation of analytical concordance in mutation-positive specimens rather than a full diagnostic performance study.

Conclusion

In this large real-world NSCLC cohort, ddPCR demonstrated high analytical concordance with NGS for EGFR mutation detection and VAF quantification across multiple specimen types, achieving an overall PPA of 98.82%. ddPCR maintained reliable performance across heterogeneous clinical materials, including fresh tumor tissue DNA, FFPE DNA, plasma cfDNA, pleural effusion supernatant DNA, and corresponding pre-capture NGS libraries. The high concordance between native DNA and pre-capture library material supports the use of archived libraries as alternative templates in low-input scenarios. When applied to predefined clinically relevant mutations, ddPCR provides a practical complement to NGS, particularly in settings requiring high analytical sensitivity, rapid turnaround, or repeated measurements across diverse sample types.